US9524319B2 - Search relevance - Google Patents

Abstract

The present invention extends to methods, systems, and computer program products for improving search relevance. In one implementation, a method identifies a first group of mappings generated for dominant queries associated with a query log. The method also identifies a second group of mappings that includes mappings between a first set of queries and categories shown for the first set of queries, and mappings between a second set of queries and clicked products for the second set of queries. Upon receipt of a query, the method applies the second group of mappings for the received query if the category mappings for the received query are in the second group of mappings. Alternatively, the method applies the first group of mappings for the received query if the category mappings for the received query are in the first group of mappings but not in the second group of mappings.

Description

CROSS-REFERENCE TO RELATED APPLICATIONS

This application is related to U.S. application Ser. No. 14/194,021, filed Feb. 28, 2014, which is incorporated herein by reference for all purposes. This application is also a continuation-in-part of U.S. application Ser. No. 14/186,717, filed Feb. 21, 2014, which is incorporated herein by reference for all purposes. That application is a continuation-in-part of U.S. application Ser. No. 13/874,023, filed Apr. 30, 2013, which is incorporated herein by reference for all purposes.

TECHNICAL FIELD

The present invention relates generally to handling search queries, and, more particularly, to improving search relevance of a query using a query classification model.

BACKGROUND

Since the advent of the Internet, many retail stores offer the option to purchase items “online” through a retail store website. With the presence of an Internet connection, consumers can direct a web browser to a retail store website by entering a Uniform Resource Locator (URL) in the address bar of the web browser. The displayed retail store website allows consumers to see the items that are available from that store, almost as if the consumer was physically in the store looking at the products that are available on the shelves. The store website may organize the items that are available from the store into different sections, categories, or departments to help facilitate the consumer navigating through the store website. Furthermore, the store website may advertise any specials that are currently occurring in an effort to entice the consumer to purchase items that are on sale.

As the consumer navigates through the website and selects a particular product, the website may display additional details about the product. For example, the website may display the retail price of the item and any discounts or sale prices that may be available. Information may be displayed about the product specifications, user reviews of the product, and an option to compare selected products to each other.

Finally, if the consumer decides to purchase a particular item, the website provides an option to add the item to a purchase queue, commonly labeled as “cart.” The cart simulates a shopping cart and allows the consumer to accumulate items from the website until they are ready to execute a transaction, pay for the products that have been added to their cart, and provide billing and shipping details.

A common difficulty for online consumers is navigating to the correct location to find the products that they are interested in researching or purchasing; especially if the product is unique or needs to be from a specific manufacturer. As more and more products become readily available for purchase, this task becomes increasingly difficult to filter out the product of interest from the innumerable other products that are available for purchase.

In order to help facilitate the online shopping experience for a customer, retail stores provide search features on their websites. The search feature allows a consumer to execute queries on product names and/or merchandise categories. Queries enable the consumer to find the products that they are interested in purchasing and/or researching in a more convenient and timely fashion. In response to a query on a particular search term, a website can return the products that most closely resemble the search terms entered by the consumer. The products are often returned in the form of a list.

Given the many products that are available for purchase over the Internet, it becomes incumbent for a retail store to optimize their search feature such that a consumer can find items of interest in a timely and efficient manner. If a retail store's search feature is not optimized, and returns results that are not of interest to the consumer, the consumer may decide to give up and not make the purchase they had intended, or to visit a different store's website. These actions can result in a loss of business to the retail store, and may serve as a deterrent, causing the consumer to not visit the website again in the future.

In order to improve the search feature's search results, many retail stores incorporate human input in addition to the search algorithms that are already present. The human input is used to modify a product's fields so that more relevant items are returned when a query is executed. However, human input requires significant effort, and is error prone.

BRIEF DESCRIPTION OF THE DRAWINGS

The specific features, aspects and advantages of the present invention will become better understood with regard to the following description and accompanying drawings where:

FIG. 1 illustrates an example block diagram of a computing device.

FIG. 2 illustrates an example computer architecture for classifying e-commerce queries to generate category mappings for dominant products.

FIG. 3 illustrates a flow chart of an example method for classifying e-commerce queries to generate category mappings for dominant products.

FIG. 4 illustrates example equations for assigning category types.

FIG. 5 illustrates an example computer architecture for classifying e-commerce queries to generate category mappings for queries with sufficient clicks.

FIG. 6 illustrates a flow chart of an example method for classifying e-commerce queries to generate category mappings for queries with sufficient clicks.

FIG. 7 illustrates example equations for assigning category types.

FIG. 8 illustrates a flow chart of an example method for training a classification model to predict categories for queries.

FIG. 9 illustrates a flow chart of an example method for improving search relevance of a query.

DETAILED DESCRIPTION

The present invention extends to methods, systems, and computer program products for improving search relevance. In a particular embodiment, a method identifies a first group of mappings generated for dominant queries associated with a query log containing a plurality of user queries. The method also identifies a second group of mappings that includes mappings between a first set of queries and categories shown for the first set of queries, and mappings between a second set of queries and clicked products for the second set of queries. For a received query, the method applies the second group of mappings for the received query if the category mappings for the received query are in the second group of mappings. Additionally, the method applies the first group of mappings for the received query if the category mappings for the received query are in the first group of mappings but not in the second group of mappings.

Embodiments of the present invention may comprise or utilize a special purpose or general-purpose computer including computer hardware, such as, for example, one or more processors and system memory, as discussed in greater detail below. Embodiments within the scope of the present invention also include physical and other computer-readable media for carrying or storing computer-executable instructions and/or data structures. Such computer-readable media can be any available media that can be accessed by a general purpose or special purpose computer system. Computer-readable media that store computer-executable instructions are computer storage media (devices). Computer-readable media that carry computer-executable instructions are transmission media. Thus, by way of example, and not limitation, embodiments of the invention can comprise at least two distinctly different kinds of computer-readable media: computer storage media (devices) and transmission media.

Computer storage media (devices) includes RAM, ROM, EEPROM, CD-ROM, solid state drives (“SSDs”) (e.g., based on RAM), Flash memory, phase-change memory (“PCM”), other types of memory, other optical disk storage, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store desired program code means in the form of computer-executable instructions or data structures and which can be accessed by a general purpose or special purpose computer.

A “network” is defined as one or more data links that enable the transport of electronic data between computer systems and/or modules and/or other electronic devices. When information is transferred or provided over a network or another communications connection (either hardwired, wireless, or a combination of hardwired or wireless) to a computer, the computer properly views the connection as a transmission medium. Hardwired connections can include, but are not limited to, wires with metallic conductors and/or optical fibers. Transmissions media can include a network and/or data links which can be used to carry desired program code means in the form of computer-executable instructions or data structures and which can be accessed by a general purpose or special purpose computer. Combinations of the above should also be included within the scope of computer-readable media.

Further, upon reaching various computer system components, program code means in the form of computer-executable instructions or data structures can be transferred automatically from transmission media to computer storage media (devices) (or vice versa). For example, computer-executable instructions or data structures received over a network or data link can be buffered in RAM within a network interface module (e.g., a “NIC”), and then eventually transferred to computer system RAM and/or to less volatile computer storage media (devices) at a computer system. RAM can also include solid state drives (SSDs or PCIx based real time memory tiered Storage, such as FusionIO). Thus, it should be understood that computer storage media (devices) can be included in computer system components that also (or even primarily) utilize transmission media.

Computer-executable instructions comprise, for example, instructions and data which, when executed at a processor, cause a general purpose computer, special purpose computer, or special purpose processing device to perform a certain function or group of functions. The computer executable instructions may be, for example, binaries, intermediate format instructions such as assembly language, or even source code. Although the subject matter has been described in language specific to structural features and/or methodological acts, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to the described features or acts described above. Rather, the described features and acts are disclosed as example forms of implementing the claims.

Those skilled in the art will appreciate that the invention may be practiced in network computing environments with many types of computer system configurations, including, personal computers, desktop computers, laptop computers, message processors, handheld devices, multi-processor systems, microprocessor-based or programmable consumer electronics, network PCs, minicomputers, mainframe computers, mobile telephones, PDAs, tablets, pagers, routers, switches, various storage devices, and the like. The invention may also be practiced in distributed system environments where local and remote computer systems, which are linked (either by hardwired data links, wireless data links, or by a combination of hardwired and wireless data links) through a network, both perform tasks. In a distributed system environment, program modules may be located in both local and remote memory storage devices.

Embodiments of the invention can also be implemented in cloud computing environments. In this description and the following claims, “cloud computing” is defined as a model for enabling ubiquitous, convenient, on-demand network access to a shared pool of configurable computing resources (e.g., networks, servers, storage, applications, and services) that can be rapidly provisioned via virtualization and released with minimal management effort or service provider interaction, and then scaled accordingly. A cloud model can be composed of various characteristics (e.g., on-demand self-service, broad network access, resource pooling, rapid elasticity, measured service, etc.), service models (e.g., Software as a Service (SaaS), Platform as a Service (PaaS), Infrastructure as a Service (IaaS), and deployment models (e.g., private cloud, community cloud, public cloud, hybrid cloud, etc.). Databases and servers described with respect to the present invention can be included in a cloud model.

Further, where appropriate, functions described herein can be performed in one or more of: hardware, software, firmware, digital components, or analog components. For example, one or more application specific integrated circuits (ASICs) can be programmed to carry out one or more of the systems and procedures described herein. Certain terms are used throughout the following description and Claims to refer to particular system components. As one skilled in the art will appreciate, components may be referred to by different names. This document does not intend to distinguish between components that differ in name, but not function.

FIG. 1 illustrates an example block diagram of a computing device100. Computing device100 can be used to perform various procedures, such as those discussed herein. Computing device100 can function as a server, a client, or any other computing entity. Computing device100 can perform various communication and data transfer functions as described herein and can execute one or more application programs, such as the application programs described herein. Computing device100 can be any of a wide variety of computing devices, such as a mobile telephone or other mobile device, a desktop computer, a notebook computer, a server computer, a handheld computer, tablet computer and the like.

Computing device100 includes one or more processor(s)102, one or more memory device(s)104, one or more interface(s)106, one or more mass storage device(s)108, one or more Input/Output (I/O) device(s)110, and adisplay device130 all of which are coupled to abus112. Processor(s)102 include one or more processors or controllers that execute instructions stored in memory device(s)104 and/or mass storage device(s)108. Processor(s)102 may also include various types of computer storage media, such as cache memory.

Memory device(s)104 include various computer storage media, such as volatile memory (e.g., random access memory (RAM)114) and/or nonvolatile memory (e.g., read-only memory (ROM)116). Memory device(s)104 may also include rewritable ROM, such as Flash memory.

Mass storage device(s)108 include various computer storage media, such as magnetic tapes, magnetic disks, optical disks, solid state memory (e.g., Flash memory), and so forth. As depicted inFIG. 1, a particular mass storage device is a hard disk drive124. Various drives may also be included in mass storage device(s)108 to enable reading from and/or writing to the various computer readable media. Mass storage device(s)108 includeremovable media126 and/or non-removable media.

I/O device(s)110 include various devices that allow data and/or other information to be input to or retrieved from computing device100. Example I/O device(s)110 include cursor control devices, keyboards, keypads, barcode scanners, microphones, monitors or other display devices, speakers, printers, network interface cards, modems, cameras, lenses, CCDs or other image capture devices, and the like.

Display device130 includes any type of device capable of displaying information to one or more users of computing device100. Examples ofdisplay device130 include a monitor, display terminal, video projection device, and the like.

Interface(s)106 include various interfaces that allow computing device100 to interact with other systems, devices, or computing environments as well as humans. Example interface(s)106 can include any number ofdifferent network interfaces120, such as interfaces to personal area networks (PANs), local area networks (LANs), wide area networks (WANs), wireless networks (e.g., near field communication (NFC), Bluetooth, and Wi-Fi networks), and the Internet. Other interfaces include user interface118 andperipheral device interface122.

Bus112 allows processor(s)102, memory device(s)104, interface(s)106, mass storage device(s)108, and I/O device(s)110 to communicate with one another, as well as other devices or components coupled tobus112.Bus112 represents one or more of several types of bus structures, such as a system bus, PCI bus, IEEE 1394 bus, USB bus, and so forth.

FIG. 2 illustrates anexample computer architecture200 for classifying e-commerce queries to generate category mappings for dominant products. Referring toFIG. 2,computer architecture200 includesquery classification module201 andquery log210. Each ofquery classification module201 and query log210 as well as their respective components can be connected to one another over (or be part of) a network, such as, for example, a PAN, a LAN, a WAN, and even the Internet. Accordingly,query classification module201 and query log210 as well as any other connected computer systems and their components, can create message related data and exchange message related data (e.g., near field communication (NFC) payloads, Bluetooth packets, Internet Protocol (IP) datagrams and other higher layer protocols that utilize IP datagrams, such as, Transmission Control Protocol (TCP), Hypertext Transfer Protocol (HTTP), Simple Mail Transfer Protocol (SMTP), etc.) over the network.

Generally, query log210 can include a plurality of query records. The query records can be accumulated withinquery log210 in response to customers entering queries against a product database, such as, for example, on an e-commerce website. As depicted,query log210 includes a plurality of query records includingquery records211,221, and231. Each query record can include data corresponding to a customer entered query. Query records can include one or more categories used as search terms, results returned in response to the one or more categories (e.g., any products that matched the search terms), click through information (e.g., product impression that were selected by a user), and a time stamp.

For example,query record211 includescategories212,results213, click throughinformation214, andtime stamp216. Similarly,query record221 includescategories222,results223, click throughinformation224, andtime stamp226. Likewise,query record231 includescategories232,results233, click throughinformation234, andtime stamp236.

Query records can also include other information, such as, for example, all of the products shown to the user, whether or not the product was added to the cart, whether or not the product was ordered, the order number, the product's primary and other category mappings, and the product position in the search results.

As depicted,query classification module201 includesrecord access module202,selection rate calculator203,product selector204,category score calculator206, and rankingmodule207.Record access module202 is configured to mine query log210 for query records with click through information that indicates one or more products were selected from among corresponding query results. As such,record access module202 can form a set of one or more categories from which a product was selected. Record access module can be configured to mine query log210 for query records within a specified time frame (e.g., within the last six months).

Selection rate calculator203 is configured to calculate a selection rate for any product returned in a query result for a specified category.Product selector204 is configured to identify a specified top number (e.g., top ten) of products in the specified category. The specified top number of products can have higher selection rates relative to other products in the specified category.Category score calculator206 is configured to calculate a category score based on product information associated with the specified top number of products in the specified category.

In some embodiments, the functionality ofselection rate calculator203,product selector204, andcategory score calculator206 are implemented on each category in set of one or more categories formed byrecord access module202. For each category,selection rate calculator203 can calculate a selection rate for any product contained in request results for the category. For each category,product selector204 can identify a specified top number of products. For each category,category score calculator206 can calculate category score for the category based on product information associated with the specified top number of products in the category.

Ranking module207 is configured to rank one or more categories relative to one another based on calculated category scores. In some embodiments, categories are assigned to different category types based on calculated category scores.

FIG. 3 illustrates a flow chart of anexample method300 for classifying e-commerce queries to generate category mappings for dominant products.Method300 will be described with respect to the components ofcomputer architecture200.

Method300 includes mining the query log for any query records with click through information that indicates one or more products were selected from among corresponding query results and that are within a specified date range (act301).Record access module202 can mine query log210 for query records with click through information that indicates one or more products were selected from among corresponding query results and that are withindate range208. For example,record access module202 can determine that click throughinformation214 indicates that one or more products were selected from amongresults213 and thattime stamp216 is withindate range208. Likewise,record access module202 can determine that click throughinformation234 indicates that one or more products were selected from amongresults233 and thattime stamp236 is withindate range208. Similar determinations can be made for other query records inquery log210.

Each ofcategories212 and232 can include one or more categories (e.g., electronics, sports, tablets, etc.) that were used as search terms in a user query of a product database.

For each of one or more categories selected from among a plurality of categories,method300 includes, calculating a selection rate for any product selected from among at least one corresponding query result returned in response to a query of the category (302). For each category incategories212 and232, selectionrate calculation module203 can calculate a selection rate for any product selected from among at least one query result returned in response to a query for the category. For example,selection rate calculator203 can calculateselection rate242 for a product selected from among at least one query result returned in response to a query for category241 (which can be a category incategories212 or232).Selection rate calculator203 can also calculateselection rate243 for another different product selected from among at least one query result returned in response to a query forcategory241.Selection rate calculator203 can also calculate selection rates for further other products selected from among at least one query result returned in response to a query forcategory241.

Similarly,selection rate calculator203 can calculate selection rates for one or more products selected from among at least one query result returned in response to a query for category251 (which can also be a category incategories212 or232). Selection rates can also be calculated for one or more products selected from among at least one query result returned in response to a for other categories incategories212 and232.

Calculating a selection rate for a product can include calculating a click-through rate based on the number of times a product was shown to users (i.e., the number of impressions) and the number of times the product was clicked on by users. Other information can also be considered when calculating selection rate for a product, including but not limited to: add to (electronic shopping) cart ratio, order ratio, and product position signals.

In some embodiments, prior to calculating selection rate, one or more products are qualified from among a plurality of products. The plurality of products is selected from among at least one corresponding query result returned in response to a query of the category. The one or more products are qualified by having one or more of: a minimum number of clicks (e.g., 2) and a minimum number of impressions (e.g., 10). In these embodiments, selection rates may not be calculated for non-qualified products.

For each of one or more categories selected from among the plurality of categories,method300 includes identifying a specified top number of products in the category, the specified top number of products having higher selection rates relative to other products in the category (303). For example,product selector204 can identify top products244 (e.g., the top ten products) incategory241.Top products242 can have higher selection rates relative to other products incategory241. Similarly,product selector204 can identify the top products (e.g., the top ten products) incategory251. The top products incategory251 can have higher selection rates relative to other products incategory251.Product selector204 can also identify the top products for other categories incategories212 and232. These other top products can have higher selection rates relative to other products in their respective categories.

For each of one or more categories selected from among the plurality of categories, calculating a category score for the category based on product information associated with the specified top number of products in the category (304). For example,category score calculator206 can calculate category score246 based on product information associated withtop products244. Similarly,category score calculator206 can calculate category score256 based on product information for the top products incategory251.Category score calculator206 can also calculate category scores for other categories incategories212 and232 based on product information for the top products in those categories respectively.

In some embodiments, a confidence interval can be used to remove bias from category score calculations. The formula used for confidence interval treatment can be varied.

It may be that the relative age of query records is considered when calculating category scores. Newer query records can be weighted to impact query score calculations more significantly. On the other hand, older query records can be weight to impact query score calculations less significantly. In other embodiments, query records are equally weighted.

In some embodiments, prior to calculating category scores, one or more categories are qualified from among a plurality of categories. A category can be qualified as a candidate when the category has a specified number of impressions (e.g., 100) or a specified period of time (e.g., six months). In these embodiments, category scores may not be calculated for non-qualified categories.

Method300 includes ranking the one or more categories based on the calculated category scores (305). For example, rankingmodule207 can rankcategory241,category251, and other categories incategories212 and232 based oncategory score246,category score256, and category scores for other categories incategories212 and232. Category rankings can be represented incategory rankings209.

In some embodiments,query classification module201 can consider the contents of (possibly many) additional query records inquery log210 when ranking categories. As described, query records can also include other information, such as, for example, all of the products shown to the user, whether or not the product was added to the cart, whether or not the product was ordered, the order number, the product's primary and other category mappings, and the product position in the search results. This other information can be considered when classifying e-commerce queries.

Products can be assigned to a plurality of categories. In some embodiments, a product is assigned to a primary category. The primary category can then be used for classifying queries. In other embodiments, each of a plurality of categories can be used for classifying e-commerce queries.

Categories can also be assigned types based on category scores.FIG. 4 illustrates example equations for assigning category types. As depicted inFIG. 4, a category can be assigned a type (e.g., 1, 2, 3, or 4) based on a calculated category score for the category.Equation402 is an example of an equation for calculating a category score. As depicted inequation402, a category score for a category can be calculated from a click-through rate for products in the category.

Equation403 is an example of an equation for calculating a click-through rate for a category. As depicted inequation403, a click-through rate for a category is based on click-through rates for products in the category.Equation404 is an example of an equation for calculating the click-through rate for a product. As depicted inequation404, a click-through rate for a product is based on product clicks and product page views within a date range.Equation406 is an example of an equation for calculating page views within a date range.Equation407 is an example of an equation for calculating clicks within a date range.Equation408 defines product page views andequation409 defines product clicks.

FIG. 5 illustrates anexample computer architecture500 for classifying e-commerce queries to generate category mappings for queries with sufficient clicks. Referring toFIG. 5,computer architecture500 includesquery classification module501 andquery log510. Each ofquery classification module501 and query log510 as well as their respective components can be connected to one another over (or be part of) a network, such as, for example, a PAN, a LAN, a WAN, and even the Internet. Accordingly,query classification module501 and query log510 as well as any other connected computer systems and their components, can create message related data and exchange message related data (e.g., near field communication (NFC) payloads, Bluetooth packets, Internet Protocol (IP) datagrams and other higher layer protocols that utilize IP datagrams, such as, Transmission Control Protocol (TCP), Hypertext Transfer Protocol (HTTP), Simple Mail Transfer Protocol (SMTP), etc.) over the network.

Generally, query log510 can include a plurality of query records. The query records can be accumulated withinquery log510 in response to customers entering queries against a product database, such as, for example, on an e-commerce website. As depicted,query log510 includes a plurality of query records includingquery records511,521, and531. Each query record can include data corresponding to a customer entered query. Query records can include one or more categories used as search terms, results returned in response to the one or more categories (e.g., any products that matched the search terms), click through information (e.g., product impression that were selected by a user), and a time stamp.

For example,query record511 includescategories512,results513, click throughinformation514, andtime stamp516. Similarly,query record521 includescategories522, results523, click throughinformation524, andtime stamp526. Likewise,query record531 includescategories532,results533, click throughinformation534, andtime stamp536.

Query records can also include other information, such as, for example, all of the products shown to the user, whether or not the product was added to the cart, whether or not the user clicked on (or otherwise selected) the product, whether or not the product was ordered, the order number, the product's primary and other category mappings, and the product position in the search results.

As depicted,query classification module501 includesrecord access module502,selection rate calculator503,product selector504,category score calculator506, and rankingmodule507.Record access module502 is configured to mine query log510 for query records with click through information that indicates one or more products were selected from among corresponding query results. As such,record access module502 can form a set of one or more categories from which a product was selected. Record access module can be configured to mine query log510 for query records within a specified time frame (e.g., within the last six months).

Selection rate calculator503 is configured to calculate a selection rate for any product returned in a query result for a specified category.Product selector504 is configured to identify a specified top number (e.g., top ten) of products in the specified category. The specified top number of products can have higher selection rates relative to other products in the specified category.Product selector504 is also configured to identify products displayed but not clicked (or otherwise selected) by the user.Category score calculator506 is configured to calculate a category score based on the number of times a category is shown and the number of times a category is clicked.

In some embodiments, the functionality ofselection rate calculator503,product selector504, andcategory score calculator506 are implemented on each category in a set of one or more categories formed byrecord access module502. For each category,selection rate calculator503 can calculate a selection rate for any product contained in request results for the category. For each category,product selector504 can identify a specified top number of products. For each category,category score calculator506 can calculate category score for the category based on product information associated with the specified top number of products in the category.

Ranking module507 is configured to rank one or more categories relative to one another based on calculated category scores. In some embodiments, categories are assigned to different category types based on calculated category scores.

FIG. 6 illustrates a flow chart of anexample method600 for classifying e-commerce queries to generate category mappings for queries with sufficient clicks.Method600 will be described with respect to the components ofcomputer architecture500.

Method600 includes mining the query log for any query records with shown product information and click through information that indicates one or more products were selected from among corresponding query results and that are within a specified date range (act601).Record access module502 can mine query log510 for query records with displayed product information and click through information that indicates one or more products were displayed and/or selected from among corresponding query results and that are withindate range508. For example,record access module502 can determine that click throughinformation514 indicates that one or more products were selected from amongresults513 and thattime stamp516 is withindate range508. Likewise,record access module502 can determine that click throughinformation534 indicates that one or more products were selected from amongresults533 and thattime stamp536 is withindate range508. Similar determinations can be made for other query records inquery log510.

Each ofcategories512,522, and532 can include one or more categories (e.g., electronics, sports, tablets, etc.) that were used as search terms in a user query of a product database.

For each of one or more categories selected from among a plurality of categories,method600 includes, calculating a product display rate and a selection rate for any product selected from among at least one corresponding query result returned in response to a query of the category (602). For each category incategories512,522, and532, selectionrate calculation module503 can calculate a selection rate (and/or a product display rate) for any product selected from among at least one query result returned in response to a query for the category. For example,selection rate calculator503 can calculateselection rate542 for a product selected from among at least one query result returned in response to a query for category541 (which can be a category incategories512,522, or532).Selection rate calculator503 can also calculateselection rate543 for another different product selected from among at least one query result returned in response to a query forcategory541.Selection rate calculator503 can also calculate selection rates for further other products selected from among at least one query result returned in response to a query forcategory541. Additionally,selection rate calculator503 can calculate a product display rate for products and/or categories.

Similarly,selection rate calculator503 can calculate selection rates for one or more products selected from among at least one query result returned in response to a query for category551 (which can also be a category incategories512,522, or532). Selection rates can also be calculated for one or more products selected from among at least one query result returned in response to other categories incategories512,522, and532.

Calculating a selection rate for a product can include calculating a click-through rate based on the number of times a product was shown to users (i.e., the number of impressions) and the number of times the product was clicked on by users. Calculating the product display rate can include calculating the number of times a product was shown to the users, regardless of whether the user clicked (or otherwise selected) the displayed product. Other information can also be considered when calculating selection rate for a product, including but not limited to: add to (electronic shopping) cart ratio, order ratio, and product position signals.

In some embodiments, prior to calculating the selection rate, one or more products are qualified from among multiple products. The multiple products are selected from among at least one corresponding query result returned in response to a query of the category. The one or more products are qualified by having one or more of: a minimum number of clicks (e.g., 2) and a minimum number of impressions (e.g., 10). In these embodiments, selection rates may not be calculated for non-qualified products.

Method600 also determines a category associated with each product (603). For each of one or more categories selected from among the plurality of categories,method600 includes identifying a mapping between a query and the shown categories, and identifying a mapping between the query and the clicked products/categories (604). For example,product selector504 can identify displayed and/or clicked products544.Top products542 can have higher click rates relative to other products incategory541. Similarly,product selector504 can identify the top displayed products (e.g., products most frequently displayed to users) incategory551.Product selector504 can also identify the most displayed and/or most clicked products for other categories incategories512,522, and532. These other top products can have higher selection rates relative to other products in their respective categories.

For each of one or more categories selected from among the multiple categories,method600 calculates a category score for the category based on the number of times a category is shown to a user and the number of times the category (or a product associated with the category) is clicked (605). For example,category score calculator506 can calculate category score546 based on product information associated with displayed and/or clicked products544. Similarly,category score calculator506 can calculate category score556 based on product information associated with displayed and/or clicked products incategory551.Category score calculator506 can also calculate category scores for other categories incategories512,522, and532 based on product information for identified products in those categories respectively.

In some embodiments, a confidence interval can be used to remove bias from category score calculations. The formula used for confidence interval treatment can be varied.

In some embodiments, the relative age of query records is considered when calculating category scores. Newer query records can be weighted to impact query score calculations more significantly. On the other hand, older query records can be weight to impact query score calculations less significantly. In other embodiments, all query records are equally weighted.

In some embodiments, prior to calculating category scores, one or more categories are qualified from among a plurality of categories. A category can be qualified as a candidate when the category has a specified number of impressions (e.g., 100) or a specified period of time (e.g., six months). In these embodiments, category scores may not be calculated for non-qualified categories.

Method600 includes ranking the one or more categories based on the calculated category scores (606). For example, rankingmodule507 can rankcategory541,category551, and other categories incategories512,522, and532 based oncategory score546,category score556, and category scores for other categories incategories512,522, and532. Category rankings can be represented incategory rankings509.

In some embodiments,query classification module501 can consider the contents of (possibly many) additional query records inquery log510 when ranking categories. As described, query records can also include other information, such as, for example, all of the products shown to the user, whether or not the user clicked (or otherwise selected) the product, whether or not the product was added to the cart, whether or not the product was ordered, the order number, the product's primary and other category mappings, and the product position in the search results. This other information can be considered when classifying e-commerce queries.

Products can be assigned to multiple categories. In some embodiments, a product is assigned to a primary category. The primary category can then be used for classifying queries. In other embodiments, multiple categories can be used for classifying e-commerce queries.

Categories can also be assigned types based on category scores.FIG. 7 illustrates example equations for assigning category types.Equation701 is an example of an equation for calculating a category score. As depicted inFIG. 7 (at702), a category can be assigned a type (e.g., 1, 2, 3, or 4) based on a calculated category score for the category.

Equation703 is an example of an equation for calculating a score for a particular category. In the first term ofequation703, “4” is added to the denominator to avoid a divide-by-zero situation. Additionally, “2” is added to the numerator to avoid a zero result in the first term ofequation703. The second term ofequation703 represents a confidence interval and is subtracted from the first term to remove any bias in the score. As depicted inequation703, a score for a category is based on the number of clicks and page views in the category.

Equation704 is an example of an equation for calculating page views for a category. As depicted inequation704, page views are considered within a date range such that the weight assigned to a particular page view decays as the date moves further into the past. For example, recent page views are given more weight inequation704 than older page views.

Equation705 is an example of an equation for calculating clicks for a category. As depicted inequation705, clicks are considered within a date range such that the weight assigned to a particular click decays as the date moves further into the past. For example, recent clicks are given more weight inequation705 than older clicks.Equation706 defines category views andequation707 defines category clicks.Equation708 is an example of an equation for calculating a maximum value for scores in a particular category. The results ofequation708 are used inequation701, discussed above.

FIG. 8 illustrates a flow chart of anexample method800 for training a classification model to predict categories for queries. Initially,method800 identifies category mappings generated for dominant queries (801). Examples of these category mappings for dominant queries are discussed herein, such as the systems and methods discussed with respect toFIGS. 2-4.Method800 also identifies mappings generated between queries and the shown categories, and identifies mappings between the queries and clicked products/categories (802). Examples of these identified mappings are discussed herein, such as the systems and methods discussed with respect toFIGS. 5-7. At803, a classification model is then trained based on all of the identified mappings (i.e., the mappings identified at801 and802).

In some embodiments, a classification model is trained using training data in the form of a “bag of words” that contains n-grams (e.g., unigrams, bigrams, etc.) associated with queries from a query log. The bag of words may also contain words obtained from product information, data entered by users, and any other data source. A “bag of words” refers to a representation of words (e.g., queries) as a bag (or multiset) without any specific relationship or ordering of the words. An n-gram is a sequence of n items (e.g., words associated with queries). In some embodiments, a unigram represents a one word query, a bigram represents a two word query, and so forth. The classification model is trained with a Naïve Bayes Multinomial model, using the n-grams in the bag of words. This model predicts categories with a score for a particular query. Although the Naïve Bayes Multinomial model is used in one embodiment, alternate embodiments may use any type of classification model.

Method800 continues by identifying additional product information from product description and other product data (804). For example, product titles, product descriptions, product features, product characteristics, and other product information can be obtained from product catalogs, online data sources, product reviews, product web sites, and the like. This additional information is useful in classifying products into one or more product categories. The additional information is also useful in associating queries with a product and/or a product category.Method800 also allows individual users to specify a category (and other information) for certain queries (805). The input from individual users is particularly useful in situations where a particular query has not previously been associated with a specific product and/or product category.

Method800 continues by updating the classification model based on the additional product information and the user input (806). In particular embodiments, the classification model is updated periodically based on updated mappings, updated product information, updated user input, and the like. Thus, the classification model is continually changing to better predict categories for received queries. Finally,method800 applies the classification model to predict categories for one or more queries (807).

FIG. 9 illustrates a flow chart of anexample method900 for improving search relevance of a query. Initially,method900 identifies category mappings generated for dominant queries and loads the mappings into a key value store (901). These identified category mappings are referred to as the “first mappings.” Examples of the category mappings for dominant queries are discussed herein, such as the systems and methods discussed with respect toFIGS. 2-4.Method900 also identifies mappings between queries and the shown categories, and identifies mappings between the queries and clicked products/categories (902). These identified mappings are referred to as “second mappings,” which are also loaded into the key value store. Examples of these identified mappings are discussed herein, such as the systems and methods discussed with respect toFIGS. 5-7.

When evaluating a query,method900 determines whether category mappings for the query are present in the second mappings (903). In this embodiment, the second mappings are considered to be the most reliable predictors of a category for the query. If category mappings for the query are present in the second mappings, the second mappings are used for the query (904). If category mappings for the query are not present in the second mappings,method900 determines whether category mappings for the query are present in the first mappings (905). If category mappings for the query are present in the first mappings, the first mappings are used for the query (906). If category mappings for the query are not present in the first mappings,method900 applies a classification model to predict categories for the query (907). An example classification model is discussed herein with respect toFIG. 8.

After determining whether to use the second mappings, the first mappings or the classification model for the query,method900 determines a category and an importance level associated with the category (908). In some embodiments, the importance level has a value of 1, 2, 3 or 4. Each importance level has an associated weight.Method900 assigns a weight to the selected category based on the importance level associated with the category (909). Alternatively, the weight value may be identified (or determined) if it was previously assigned. For example,importance level 1 is weighted higher thanimportance level 2,importance level 2 is weighted higher thanimportance level 3, andimportance level 3 is weighted higher thanimportance level 4. The weight is used in calculating a score, as discussed herein, for each product or category associated with a particular query. In some embodiments, a solr scoring function is used to calculate scores. Solr is an open source enterprise search platform. The importance levels and associated weights may be determined manually by a user or automatically by one or more systems or methods discussed herein. In one embodiment, the importance levels and associated weights are determined byquery classification module201 and/or501.

Finally, the category mappings are boosted using the assigned weight (910). For example, if a particular category has a weight of 2, the boosted value of the query is squared (BoostedCategoryValue=CategoryValue²). This boosted value may result in presenting products in the boosted category higher in the search results since these products are likely to be more relevant to the query.

Although the components and modules illustrated herein are shown and described in a particular arrangement, the arrangement of components and modules may be altered to process data in a different manner. In other embodiments, one or more additional components or modules may be added to the described systems, and one or more components or modules may be removed from the described systems. Alternate embodiments may combine two or more of the described components or modules into a single component or module.

The foregoing description has been presented for the purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise form disclosed. Many modifications and variations are possible in light of the above teaching. Further, it should be noted that any or all of the aforementioned alternate embodiments may be used in any combination desired to form additional hybrid embodiments of the invention.

Further, although specific embodiments of the invention have been described and illustrated, the invention is not to be limited to the specific forms or arrangements of parts so described and illustrated. The scope of the invention is to be defined by the claims appended hereto, any future claims submitted here and in different applications, and their equivalents.

Claims (20)

What is claimed:

1. A method of improving relevance of products identified in search results provided in response to product queries entered by consumers in search of products, the method comprising:

identifying a first group of product category mappings of dominant product queries in a query log containing a plurality of consumer product queries;

identifying a second group of mappings between the consumer product queries and shown product categories that includes mappings between a first set of the consumer product queries and product categories of the first set of the consumer product queries, and mappings between a second set of the consumer product queries and clicked products of the second set of the consumer product queries;

receiving a new consumer product query;

using one or more processors, evaluating whether category mappings of the new consumer product query are in the second group of mappings;

when the category mappings of the new consumer product query are in the second group of mappings, considering the second group of mappings to predict a category of the new consumer product query; and

when the category mappings of the new consumer product query are in the first group of product category mappings but not in the second group of mappings, considering the first group of product category mappings to predict the category of the new consumer product query.

2. The method ofclaim 1 further comprising applying a classification model to predict the category of the new consumer product query when:

the category mappings of the new consumer product query are not in the first group of product category mappings; and

the category mappings of the new consumer product query are not in the second groups of mappings.

3. The method ofclaim 1 further comprising:

presenting search results of the new consumer product query based on the first group of product category mappings when the first group of product category mappings has been considered to predict the category of the new consumer product query; and

presenting search results of the new consumer product query based on the second group of mappings when the second group of mappings has been considered to predict the category of the new consumer product query.

4. The method ofclaim 1 further comprising:

loading the first group of product category mappings into a key value store; and

loading the second group of mappings into the key value store.

5. The method ofclaim 4 further comprising determining an importance level associated with the category of the new consumer product query.

6. The method ofclaim 5 further comprising determining a weight associated with the category of the new consumer product query based on the importance level.

7. The method ofclaim 6 further comprising presenting a product higher in the search results based on the weight.

8. The method ofclaim 1 further comprising presenting the search results to the consumers.

9. A method comprising:

receiving a product query;

identifying a first group of mappings of a first group of dominant product queries in a query log;

identifying a second group of mappings that includes mappings between a second group of product queries in the query log and clicked products of the second group of product queries in the query log;

applying, using one or more processors, the second group of mappings to the product query when category mappings for the product query are in the second group of mappings;

applying, using the one or more processors, the first group of mappings for the product query when:

the category mappings for the product query are in the first group of mappings; and

the category mappings for the product query are not in the second group of mappings; and

applying, using the one or more processors, a classification model to predict categories for the product query when:

the category mappings for the product query are not in the first group of mappings; and

the category mappings for the product query are not in the second group of mappings.

10. The method ofclaim 9 further comprising presenting search results for the product query based on the first group of mappings, the second group of mappings, or the classification model.

11. The method ofclaim 9 further comprising determining a product category associated with the product query.

12. The method ofclaim 11 further comprising determining an importance level associated with the product category.

13. The method ofclaim 12 further comprising determining a weight associated with the product category based on the importance level.

14. The method ofclaim 13 further comprising boosting a category mapping based on the weight.

15. The method ofclaim 14 further comprising presenting search results for the product query based on the category mapping that was boosted based on the weight.

16. An apparatus comprising:

a memory device; and

one or more processors operably coupled to the memory device, the one or more processors configured to:

receive a product query from a customer;

identify a first group of mappings generated for dominant product queries from a query log containing product queries made by previous customers;

identify a second group of mappings from the query log that includes mappings between a first set of product queries from the query log and categories shown in the query log for the first set of product queries, and mappings between a second set of product queries from the query log and clicked products from the query log for the second set of product queries;

apply the second group of mappings for the product query from the customer when category mappings for the product query from the customer are in the second group of mappings; and

apply the first group of mappings for the product query from the customer when the category mappings for the product query from the customer are in the first group of mappings but not in the second group of mappings.

17. The apparatus ofclaim 16 wherein the one or more processors are further configured to apply a classification model to predict categories for the product query from the customer when:

the category mappings for the product query from the customer are not in the first group of mappings; and

the category mappings for the product query from the customer are not in the second group of mappings.

18. The apparatus ofclaim 16 wherein the one or more processors are further configured to present search results for the product query from the customer based on:

the first group of mappings when the category mappings for the product query from the customer are in the first group of mappings but not in the second group of mappings; and

the second group of mappings when the category mappings for the product query from the customer are in the second group of mappings.

19. The apparatus ofclaim 16 wherein the one or more processors are further configured to:

determine a product category associated with the product query from the customer;

determine an importance level associated with the product category;

determine a weight associated with the product category based on the importance level; and

boost a category mapping based on the weight.

20. The apparatus ofclaim 19, the one or more processors further configured to present search results for the product query from the customer based on the category mapping.

Images